Investigation On The Performance Enhancement Of A Delay-based Reservoir Computing System By Using Gradient Boosting Technology

With the development of the times and the change of technology,people have a great demand for efficient,fast and accurate information processing methods.Photonic reservoir computing(RC)is a relatively novel information processing method who includes two categories: spatially distributed RC and delay-based RC.Although spatially distributed RC has good processing capacity,it has high cost and difficult node state collection,which is not conducive to practical implementation.In order to overcome these issues,delay-based RC systems are proposed.This structure is based on a single node and external feedback loop.Compared with the spatially distributed RC,the delay-based RC has the advantages of low cost,simple structure and easy implementation.Since the semiconductor laser has high relaxation oscillation frequency and rich dynamic behavior under delayed optical feedback,the delay-based RC system based on the semiconductor laser is of great research value.Nowadays,a great deal of researches on delay-based RC system based on semiconductor laser have been reported.At the same time,in the processing of complex tasks,the performance of delay-based RC has a lot of room for improvement.Therefore,the methods to improve the performance of delayed RC are worthy of further study.In this work,a technology to enhance the performance of the reservoir computing,namely gradient boosting technology,which has been proved to be effective in the spatially distributed RC,is introduced into the delay-based RC,and the effectiveness of this technology to improve the performance of the delaybased RC is numerically studied.In this scheme,the VCSEL is used as the nonlinear node of the reservoir,and a delay-based RC system is built on this basis.Through numerical simulation,it is verified that VCSEL can generate rich dynamic states under appropriate parameters,therefore it is very suitable for the nonlinear node of the delay-based reservoir.Meanwhile,the VCSEL can output two mutually perpendicular modes of X polarization and Y polarization under appropriate parameters,two types of feedback modes can be generated:polarization-rotated optical feedback(PR-OF)and polarization-preserved optical feedback(PP-OF).One of reservoir in the system is the VCSEL under PR-OF whose output is trained according to the target value of the task and named as the original reservoir.Another VCSEL under PR-OF or PP-OF is added as the extra reservoir whose target value of training is the error generated by the original reservoir.We expect to introduce an extra reservoir to correct the error of the original reservoir,so as to reduce the error of the task and realize the performance enhancement of the delay-based RC system.In this work,the Santa-Fe chaotic time series prediction task and 10 th order nonlinear autoregressive moving average(NARMA10)task are used as the benchmark tasks.By adjusting the injection strength and feedback strength of extra reservoir,the performances of the delay-based RC system are evaluated before and after supplementing the VCSEL-based extra reservoir,and then the effectiveness of the gradient boosting technology in the delay-based RC system can be analyzed.At the same time,we also compared the performance enhancement effect of the extra reservoir under PR-OF and PP-OF.The simulated results demonstrate that using gradient boosting technology can effectively improve the performance of delay-based RC system.Comparatively speaking,the enhanced effect is more obvious under taking a VCSEL with PR-OF as the extra reservoir.